Font Size: 

Recommendation algorithms, often rely on user-item interaction matrices, to uncover hidden patterns and preferences. These matrices play a pivotal role in facilitating the detection of matching similarities between users and items. However, these matrices do not capture the full spectrum of users’ preferences in ratings while providing a list of recommendations. Since such variability can be effectively modeled as symbolic objects, specifically histogram objects, it is proposed to use the Symbolic Data Analysis (SDA) tools to address this challenge. This inclusion of user preferences and item characteristics into histograms enhanced the user profile capabilities in our methodology. These profiles can then be compared using Wasserstein similarity measures to compute the nearness between users and items, enabling the recommender system to generate top-N relevant recommendations. To evaluate the efficacy of the proposed SDA-RecSys, experiments are conducted to assess the impact of histogram profiles on recommendations, by utilizing the Normalized Discounted Cumulative Gain (NDCG) metric as a benchmark. Comparisons are presented to project the superiority of the SDA framework for Recommendation systems.

Information Overload, Recommender Systems, Histogram Objects, Symbolic Data Analysis (SDA)